人格
面试
自动化
主题(计算)
计算机科学
人工智能
生成语法
数据科学
人机交互
工程类
设计思维
知识管理
软件工程
表(数据库)
参与者观察
设计方法
作者
Lauren Bertelsen,Ada-Rhodes Short
出处
期刊:
[American Society of Mechanical Engineers]
日期:2025-08-17
标识
DOI:10.1115/detc2025-169620
摘要
Abstract This paper investigates the use of ChatGPT for automating two common design methods, Photographic Interviews and Observations to Insights. Designers would have previously considered these methods challenging to automate due to their abstract, subjective, and deeply personal nature, but recent developments in AI have made automation of these qualities possible. In this paper, ChatGPT acts as three separate interview personas and also acts as the interviewer. The personas each “take photos” about an interview theme through generative AI, and the interviewer generates inferred observations about the photos. ChatGPT then generates insights about the persona and the interview theme from those observations. From these insights, we gain valuable knowledge about ChatGPT’s proficiency in extrapolating details about a given persona and its suitability for roleplaying as an interviewee. We find that there are several areas of inconsistency in the photographic interview images, including but not limited to age, race, and perspective. We also find that the observations and insights are all reasonable and appropriately inferred from the images. Given these findings, we conclude that while there is more roleplay preparation required for ChatGPT to sufficiently stand in for human interviewees, the system shows significant potential for the automation of human-centric design methods.
科研通智能强力驱动
Strongly Powered by AbleSci AI